ByteDance’s Seed division has restructured its framework ahead of next-generation ultra-large models with estimated parameters over 500 billion. Four new first-level departments were established under Wu Yonghui’s leadership, merging previously scattered teams to boost R&D efficiency. The Pretrain Data team handles multi-modal data and large-scale pre-training for the Omni model, while the Horizon RL team focuses on improving model intelligence via reinforcement learning. Separate departments manage B-end Agentic model integration and C-end chat models, with dedicated leaders for AI safety exploration. This shift aims to reduce overlapping efforts and improve communication after technical boundaries blurred in earlier team structures, providing a solid foundation for ByteDance’s competition in the global AI race.